Physician-Friendly Machine Learning: A Case Study with Cardiovascular Disease Risk Prediction.

Area of Science:

• Biomedical research
• Machine learning applications
• Artificial Intelligence in healthcare

Background:

• Machine learning (ML) is perceived as complex, limiting its adoption by physicians and biologists.
• A knowledge gap exists regarding the practical application of ML in clinical research.

Purpose of the Study:

• To demonstrate that auto machine learning (AutoML) techniques are accessible for biomedical researchers.
• To show AutoML can build competitive ML classifiers without deep algorithmic expertise.
• To challenge the perception of ML as a tool exclusively for experts.

Main Methods:

• Comparison of AutoML (auto-sklearn) against manual ML model development by a graduate student.
• Evaluation metrics included time to model building and classification accuracy on cardiovascular disease risk prediction.
• Experiments were conducted on two publicly available datasets.

Main Results:

• AutoML built superior ML classifiers in 1 hour compared to one month of manual effort.
• The AutoML-generated classifiers achieved better accuracy on test datasets.
• Building effective ML models with AutoML requires minimal coding.

Conclusions:

• AutoML significantly reduces the time and expertise needed for ML model development in biomedical research.
• AutoML democratizes AI, encouraging its adoption in clinical practice.
• This approach can accelerate the use of AI in predicting diseases like cardiovascular conditions.